Stochastic simulation is a useful technique for modeling results of complicated problems having many random components and interdependencies. Whether by design or not, such simulation models often become black boxes which can be problematic for acting on results. Also, imperfect convergence of the pseudo-random results, particularly at a detailed level can lead to sub-optimal conclusions. For both of these reasons, it is very helpful to build models of the simulation results themselves.

While these take the form of predictive models, it is really as an explanation, simplification, and convergence improvement that these models are useful. Some examples where such post-simulation models are useful include claim life-cycle models, catastrophe models, and DFA models. The approach will be described and we will also discuss concerns around contract issues regarding third party software where reverse engineering of models is prohibited.